Posts by vseven

1) Message boards : Number crunching : Optimizing the apps (Message 1103)
Posted 29 Sep 2018 by vseven
Post:

From what I've been reading on Optimizations here in this Thread, your 1080 and 2080 "SHOULD" be able to adjust "kernels_per_reduction='x'" to 64 without issue. I'd put "lut_size='x'" to 16, like I did for my 1050s. "sieve_size='x'" I set to 27, down from 30 - at 30 the System was becoming sluggish and hard to switch, (Alt-Tab), between Programs and Firefox had issues switching Tabs. At 27, System is stable again. How is your System at 25???


I tried setting the kernels per reduction to 64 from 48 and it had no change in overall speed but it did use more GPU memory, almost 30% more. Since it wasn't running any faster I put it back to 48. Sieve size at 30 is fine since the computer is a NAS and graphic performance doesn't matter. Honestly with it at 30 I could still do whatever I needed to do. At 31 I got computational errors. 25 was much slower. Highest I could go on threads was 7, I could go higher but the other settings (lut, sieve) had to go lower to not get errors and then my speed was worse. everything I change results in slower times so I'm thinking what I'm running right now, 171 - 175, is as good as I'll get on a 2080.

I would guess the 2080 Ti could get into the 150 range. But it was out of my price range. That being said a 2070 could probably compete with a 1080 Ti and those should be out in a couple weeks. I think I might even get two 2070's and sell off my 2080. Well...once they start making hybrid versions.
2) Message boards : Number crunching : Optimizing the apps (Message 1100)
Posted 29 Sep 2018 by vseven
Post:
Got a RTX 2080. Its performance on Amicable Numbers is over 2x my old 1080....went from 310 seconds per WU to 134. So working on Collatz I started with most of the 1080 optimizations (kernels_per_reduction=48, threads=8, lut_size=17, sieve_size=25) to see a apples to apples comparison and I went from averaging 345 seconds per WU to 317. Decent change but looking at the GPU it was sitting at 78% - 80% used now and not 99% like the 1080 was. So I tried increasing the threads to 9 and got computational errors. Dropped it back to 8 and increased the lut size. More computation errors. So I kept tweaking numbers and I'm down to 176 sec per WU, almost half the time. Gonna keep playing and see if I can do better but overall that's a huge difference.
3) Message boards : Number crunching : Validation Pending? (Message 801)
Posted 18 Jul 2018 by vseven
Post:
Here we go again!


Maybe its a windows version issue? I'm seeing no issue on my 3 hosts using GPU:

17 Jul 2018, 22:38:07 UTC 	18 Jul 2018, 12:29:50 UTC 	Completed and validated 	317.39 	2.14 	30,166.31 	Collatz Sieve v1.30 (opencl_nvidia_gpu)
windows_x86_64
17 Jul 2018, 22:38:07 UTC 	18 Jul 2018, 12:29:50 UTC 	Completed and validated 	324.45 	2.38 	28,699.09 	Collatz Sieve v1.30 (opencl_nvidia_gpu)
windows_x86_64
17 Jul 2018, 22:38:07 UTC 	18 Jul 2018, 12:29:50 UTC 	Completed and validated 	322.40 	2.08 	28,130.42 	Collatz Sieve v1.30 (opencl_nvidia_gpu)
windows_x86_64
17 Jul 2018, 22:38:07 UTC 	18 Jul 2018, 12:29:50 UTC 	Completed and validated 	321.37 	2.38 	27,938.45 	Collatz Sieve v1.30 (opencl_nvidia_gpu)
windows_x86_64
4) Message boards : Number crunching : Validation error (Message 714)
Posted 10 Jul 2018 by vseven
Post:
832118

Look when this host was created, when tasks were sent, when the results returned...


And that is one of mine. I'm running an optimized config and I have both a 1080 hybrid that's overclocked and a Titan V. My PC flys through WUs....there is no cheating going on. And I'm only partially using the Titan for this project, most of the time it's crunching Milky Way or amicable number work units.

What really sucks for me is I just noticed yesterday night. Only saving grace is the 1080 is dedicated to this project and since the Titan is doing other things I didn't lose that much.
5) Message boards : Number crunching : Optimizing the apps (Message 431)
Posted 24 May 2018 by vseven
Post:
nVidia Tesla v100 SXM2. Would imagine the Titan V to be similar and possibly the next gen cards:

verbose=1
kernels_per_reduction=48
threads=7
lut_size=19
sleep=1
reduce_cpu=0
sieve_size=30

Averaging 87 seconds per WU. Tested a server with 8 cards so I was cranking out a WU every 11 seconds. Only for 2 hours unfortunately (all the test time I had).
6) Message boards : Number crunching : Optimizing the apps (Message 375)
Posted 15 May 2018 by vseven
Post:
As I said post #71 in this thread gives all the data I know, personally I would pick one close to yours and go from there.


Is there a reason they say AMD 1070, AMD 1080, AMD 1080 ti, etc? Shouldn't those all be GTX or nVidia?




©2020 Jon Sonntag; All rights reserved